Image clustering with CNN as feature extractor

Hello,

I was wondering how I would go about to do clustering (unsupervised learning) of images using CNN (ResNet or VGGNet) as a feature extractor? I have read the guide on how to use CNN as a feature exactor and using an SVM rather than softmax layer to classify the images in the case of supervised learning. I was wondering if it was possible to do a k-means clustering using the extracted features in a similar way.

Thank you

Answer 1

[Mahesh Taparia]

Hi

I assume you want to apply the k-means clustering on the features extracted from the trained CNN.

You can perform the clustering on the feature extracted from the trained CNN. It will try to create a cluster of same categorical images. Experiment it, it may work. It will be a supervised approach as to get the trained network, you will require the label information.

But if you are having the label information, then it is advisable to go with CNN based approach instead of extracting feature and applying kmeans.